address comment

Signed-off-by: leslie-fang25 <[email protected]>

Author: leslie-fang25

tensorrt_llm/llmapi/llm_args.py

@@ -2458,9 +2458,6 @@ def validate_batch_wait_timeout_ms(self) -> 'TorchLlmArgs':
             raise ValueError("batch_wait_timeout_ms must be greater than 0")
         return self
 
-    def set_mm_encoder_only(self, mm_encoder_only):
-        self.mm_encoder_only = mm_encoder_only
-
     def get_executor_config(
         self,
         _hf_model_dir: Optional[Path] = None,

tensorrt_llm/llmapi/mm_encoder.py

@@ -56,7 +56,7 @@ def _build_model(self):
         self._tokenizer = self.input_processor.tokenizer
 
         assert isinstance(self.args, TorchLlmArgs)
-        self.args.set_mm_encoder_only(True)
+        self.args.mm_encoder_only = True
 
         self._executor = self._executor_cls.create(
             self._engine_dir,

tests/unittest/llmapi/test_llm_args.py

@@ -438,6 +438,13 @@ def test_runtime_sizes(self):
             assert llm.args.max_seq_len == 128
             assert llm.args.max_batch_size == 8
 
+            executor_config = llm.args.get_executor_config(
+                llm._hf_model_dir, llm.tokenizer)
+            assert executor_config.max_beam_width == 1
+            assert executor_config.max_num_tokens == 256
+            assert executor_config.max_seq_len == 128
+            assert executor_config.max_batch_size == 8
+
     def test_dynamic_setattr(self):
         with pytest.raises(pydantic_core._pydantic_core.ValidationError):
             args = TorchLlmArgs(model=llama_model_path, invalid_arg=1)